Autonomous vehicles are a growing technology that aims to enhance safety, accessibility, efficiency, and convenience through autonomous maneuvers ranging from lane change to overtaking. Overtaking is one of the most challenging maneuvers for autonomous vehicles, and current techniques for autonomous overtaking are limited to simple situations. This paper studies how to increase safety in autonomous overtaking by allowing the maneuver to be aborted. We propose a decision-making process based on a deep Q-Network to determine if and when the overtaking maneuver needs to be aborted. The proposed algorithm is empirically evaluated in simulation with varying traffic situations, indicating that the proposed method improves safety during overtaking maneuvers. Furthermore, the approach is demonstrated in real-world experiments using the autonomous shuttle iseAuto.

HD Maps (High-Definition Maps) serve as crucial resources for the domain of autonomous vehicle. Because HD Maps can provide detailed and accurate road information, the generation of HD Maps has been a labour-intensive and high cost. This research presents an innovative and semi-automated approach for efficient HD Maps generation by using assure mapping tool with deep learning techniques and mobile laser scanned point cloud geometry. The proposed method starts with data collection from various sources such as images, LiDAR point clouds, and integrated INS/GNSS trajectory data. These data are labelled by using a pre-trained model. After finishing post-labelling, these data are subjected to deep learning training by using VoxelNet and Yolact++ framework and leading to the generation of an AI model. The tool effectively recognizes and categorizes features such as road surface markings, traffic signs, and traffic lights, which can be further expanded as per requirements. Finally, the output format can be converted to OpenDRIVE, Lanelet2, and other else. Hence, the extracted lane lines can compare to the manual mapping data for verifying the accuracy. This study demonstrates that the proposed approach can be instrumental in streamlining the HD Maps generation procedure, reducing manual labour, and enhancing efficiency. The assure mapping tool proves to be an effective instrument, particularly when powered by deep learning algorithms and point cloud geometries, in the creation of reliable, comprehensive, and application-ready HD Maps.

Mobile mapping technologies, for example multi-sensor integration and multi-platform mapping technology, have developed and improved over recent decades, various applications such as conventional mapping scenarios, rapid disaster response, smart city, and autonomous vehicle application arise synchronously. Especially, autonomous driving vehicles have made enormous progress. High-definition (HD) maps are key for autonomous driving because of their high accuracy and rich information of road scenes. However, how to make sure that HD maps are suitable for autonomous vehicle requirement is an important topic. The HD maps guidelines and standards in Taiwan are released since 2018 and mainly focus on point cloud and shape file format. In this paper, a procedure for the verification and validation of HD maps for OpenDRIVE and Autoware (Lanelet2) is proposed. It discusses about the verification strategies, suggestion review item, recommendation tools, and process. As shown by our preliminary results, the proposed process can conform not only in closed area but also public road. These issues can help reducing HD maps production costs. When the foundation of HD maps accomplishes, the autonomous driving techniques can naturally complement. The vision of full automation vehicle will come true rapidly in the future.

Abstract. As research on autonomous driving deepens, High-definition Maps (HD Maps) have gradually become an auxiliary information for the new generation of autonomous driving technology. Compared to traditional electronic navigation maps, HD Maps have higher accuracy requirements and more information. Multi-road environment information and road elements are included. In the production of HD Maps, the on-board Mobile Laser Scanning (MLS) system has the ability to quickly collect environmental information, with high precision, thus making the system a widely used data collection method today. However, subsequent map building, digitization, and other mapping work still rely on manual operation, which is time-consuming and laborious. Therefore, this research is dedicated to developing a semi-automatic algorithm to generate HD Maps from the acquired point cloud data. This research focuses on the extraction of road surface markings, using the Cloth Simulation Filter (CSF) to obtain the road surface point cloud to improve the extraction efficiency. The road markings are extracted using the characteristic of high intensity values, and the commonly used Otsu threshold filter in image processing is used to extract point clouds with high reflectance intensity, eliminating the need for manual setting of point clouds. And based on geometric conditions, the objects are classified, such as arrow lines, pedestrian crossings, stop lines, and lane lines, which are convenient for further mapping HD Maps.

As the operational domain of autonomous vehicles expands, encountering occlusions during navigation becomes unavoidable. Most of the existing research on occlusion-aware motion planning focuses only on the longitudinal motion of the ego vehicle and neglects its lateral motion, resulting in output motion that can be overly conservative. This paper proposes a motion planner capable of actively adjusting the ego vehicle's lateral position to minimize occlusions. The proposed planner is applicable in various scenarios and can function under perception uncertainty. This work also extends our previously proposed 3D visibility estimation approach for addressing occlusions caused by objects which are not present in HD maps. The proposed planner first generates candidate trajectories. The current and future visibility of each trajectory is then estimated using live LiDAR data and HD maps. These estimated visibilities are converted into visibility costs, which are then used to determine the optimal output trajectory in conjunction with other planning costs. The proposed planner is tested in three scenarios using the CARLA simulator: an occluded T-junction crossing, turning at a low-visibility corner and preparing to pass a parked vehicle, using live localization and object detection results. The experimental results reveal that the proposed planner allows the ego vehicle to minimize occlusions by diverging from the center of the lane and, consequently, to discover occluded vehicles earlier than a baseline planner in most situations. Moreover, occlusions caused by a parked vehicle not present in the HD maps were estimated correctly using our extended visibility estimation method.

Mapping technologies have improved over time, and autonomous driving techniques have advanced substantially over recent decades. High-definition (HD) maps are key for autonomous driving because of their accurate and rich interpretations of road scenes. HD maps provide information about road features, such as lane lines, centerlines, traffic signs, and traffic lights, to help autonomous vehicles navigate safely. HD maps have three major challenges: the standardization of the format of HD maps, conversion between map formats, and lack of techniques for automated HD map generation. These issues influence the costs of HD maps. Therefore, this article proposes strategies to overcome these challenges as well as control the cost with the support of the Ministry of the Interior in Taiwan. We established relevant HD map standards and guidelines to standardize the HD map production procedure. Additionally, we contribute to developing semiautomated HD map production tool to enhance the efficiency of HD map production. Another contribution is to develop HD map format conversion tool to satisfy the map requirement for different end-user. This project not only promotes the development of the Taiwanese autonomous driving industry but also increases its international competitiveness.

High-definition (HD) mapping is a promising approach to realize highly automated driving (AD). Although HD maps can be applied to all levels of autonomy, their use is particularly beneficial for autonomy levels 4 or higher. HD maps enable AD systems to see beyond the field of view of conventional sensors, thereby providing accurate and detailed information regarding a driving environment. An HD map is typically separated into a pointcloud map for localization and a vector map for path planning. In this paper, we introduce two separate but successive HD map generation workflows. Of the several stages involved, the registration and mapping processes are essential for creating the pointcloud and vector maps, respectively. To facilitate the readers' understanding, the processes of these two stages have been recorded and uploaded online. HD maps are typically generated using open-source software (OSS) tools. CloudCompare and ASSURE, as representative tools, are used in this study. The generated HD maps are validated with localization and path-planning modules in Autoware, which is also an OSS stack for AD systems. The generated HD maps enable environmental-monitoring vehicles to successfully operate at level 4 autonomy.

There are few open source autonomous driving planners that are general enough to be used directly, or which could be easily customized to suit a particular application. OpenPlanner 1.0 was introduced back in 2017 to fill this gap. It was developed and integrated with the open source autonomous driving framework Autoware. Since then, many improvements have been introduced, following the original design goals. In this paper, the basic design will be re-introduced along with the latest developed technologies. The new planner is called OpenPlanner 2.0 and includes several new techniques such as multiple HD road network map formats support, trajectory and behavior estimation, planning based HMI support, path generation using kinematics based motion simulation and lane change behavior. OpenPlanner 2.0 is already attracting attention from the autonomous driving research and development communities; universities and companies. Several projects are using and contributing to its code base. Some of these applications will be discussed in this paper as well. A comparison between OpenPlanner and other open source planners showing the aspects where it is superior is also presented.

As the autonomy level of self-driving vehicles increases, they will be expected to operate safely in increasingly complex environments. During real-world driving, occlusions are inevitable. Therefore, the ability to accurately identify the visible and occluded regions surrounding an autonomous vehicle is crucial for safe operation. In this paper, a method for estimating visibility using 3D point clouds and road network maps is proposed. The proposed method projects the positions of the surrounding lanes, obtained from a road network map, and a 3D scan of the driving environment approximated from a point cloud map, onto depth images. The depth images are then compared in order to determine the visible and occluded regions of the driving environment from the specified viewpoint. Furthermore, a visibility ratio, which is a numerical value that encapsulates visibility information of a particular location, is proposed. The visibility ratio is calculated by dividing the visible area of interest by the total driving area relevant to that location. The proposed method was tested and found to be applicable in both simulated and real-world driving environments. Moreover, the experimental results show that the visibility ratio was representative of the actual visibility from particular locations.

Estimating the intentions and trajectories of other vehicles is critical to achieving stable, long-term planning and decision making in autonomous driving systems. This paper introduces a novel technique for estimating the intention and trajectory probabilities of surrounding vehicles. The first step is to use a deterministic behavior planner to identify possible trajectories and behaviors. The behavior planner models an average driver who follows the driving rules according to information provided by a road network map, and also provides the control signal used in the following step. Next, a customized particle filter is integrated with the planner to model the uncertainty of various trajectories and behaviors, using multiple sensing cues such as pose, velocity, acceleration and turn signal use. The proposed method supports various sensor modalities, depending on the availability of additional sensing information. Finally, by including the sensor data the probabilistic process is able to estimate the probabilities of various trajectories and intentions. The proposed method is generic to any driving situation supported by the behavior planner. Intentions such as ‘go forward’, ‘turn right’, ‘turn left’, ‘yield’ and ‘stop’ are supported by the proposed method. Our proposed method is evaluated using multiple, complex, simulated driving situations, and then comparing the simulation’s ground truth to the estimated probabilities. Evaluation criteria are how early and accurately our system can estimate the driving trajectory and intention probabilities of the vehicle. Our results show that the method can successfully estimate driver intention and trajectory in multiple complex situations, such as left turns, right turns and during passing, as well as at four-way intersections and bus stops.

In this paper, we develop a tool to generate vector map necessary for path planning in Autoware, an open-source autonomous driving software. Vector map defines road network structure, from which the global path is produced in planning module. Although Autoware is open-source, an approach to create customized vector map is considered as business nature. The free available vector mapper tool for Autoware is also not reliable and requires point cloud data file as input. However, this proposed tool takes only waypoints, locations at which a car drives, as input and produces vector map.

Autonomous mobile robot navigation in real unmodified outdoor areas frequented by people on their business, children playing, fast running bicycles, and even robots, remains a difficult challenge. For eleven years, the Tsukuba Challenge Real World Robot Challenge (RWRC) has brought together robots, researchers, companies, government, and ordinary citizens, under the same outdoor space to push forward the limits of autonomous mobile robots. For the Tsukuba Challenge 2017 participation, our team proposed to study the problem of sensors-to-actuators navigation (also called End-to-End), this is, having the robot to navigate towards the destination on a complex path, not only moving straight but also turning at intersections. End-to-End (E2E) navigation was implemented using a convolutional neural network (CNN): the robot learns how to go straight, turn left, and turn right, using camera images and trajectory data. E2E network training and evaluation was performed at Nagoya University, on similar outdoor conditions to that of Tsukuba Challenge 2017 (TC2017). Even thought E2E was trained on a different environment and conditions, the robot successfully followed the designated trajectory in the TC2017 course. Learning how to follow the road no matter the environment is of the key attributes of E2E based navigation. Our E2E does not perform obstacle avoidance and can be affected by illumination and seasonal changes. Therefore, to improve safety and add fault tolerance measures, we developed an E2E navigation approach with model-based system as backup. The model-based system is based on our open source autonomous vehicle software adapted to use on a mobile robot. In this work we describe our approach, implementation, experiences and main contributions.

This paper describes our approach to perform robust monocular camera metric localization in the dynamic environments of Tsukuba Challenge 2016. We address two issues related to vision-based navigation. First, we improved the coverage by building a custom vocabulary out of the scene and improving upon place recognition routine which is key for global localization. Second, we established possibility of lifelong localization by using previous year's map. Experimental results show that localization coverage was higher than 90% for six different data sets taken in different years, while localization average errors were under 0.2 m. Finally, the average of coverage for data sets tested with maps taken in different years was of 75%.

Planning is one of the cornerstones of autonomous robot navigation. In this paper we introduce an open source planner called "OpenPlanner" for mobile robot navigation, composed of a global path planner, a behavior state generator and a local planner. OpenPlanner requires a map and a goal position to compute a global path and execute it while avoiding obstacles. It can also trigger behaviors, such as stopping at traffic lights. The global planner generates smooth, global paths to be used as a reference, after considering traffic costs annotated in the map. The local planner generates smooth, obstacle-free local trajectories which are used by a trajectory tracker to achieve low level control. The behavior state generator handles situations such as path tracking, object following, obstacle avoidance, emergency stopping, stopping at stop signs and traffic light negotiation. OpenPlanner is evaluated in simulation and field experimentation using a non-holonomic Ackerman steering-based mobile robot. Results from simulation and field experimentation indicate that OpenPlanner can generate global and local paths dynamically, navigate smoothly through a highly dynamic environments and operate reliably in real time. OpenPlanner has been implemented in the Autoware open source autonomous driving framework's Robot Operating System (ROS).

In this paper we present the design of a car test bed. The main characteristic of our test bed is the steering which is Non-Holonomic to resemble the normal car steering. This test bed is a rear wheel drive with frontal steering by servo motor. The control brain consists of RoBoard II and Arduino Micro controller. We used multiple sensors (Ultrasonic range finder, 6 DOF Accelerometer, Wheel encoder and Stereo Camera). For communication, 150 Mbps wireless attaches to the RoBoard. We develop a C++ control framework for the basic test bed functions which includes (PID controller, Vision frame grabber and preprocessor and basic obstacle avoidance using Ultrasonic range finder). Key-Words: Test bed, Non-Holonomic, Sensors, Hardware Architecture, Software Architecture, PID controller.